Fitgrid

Bug fixes

- fix stupid bug where `lmer` did not use specified channels

Bug fixes

- add option to plot legend separately in `plot_betas`, otherwise we fail miserably with many betas
- guard against duplicate channels in `Epochs` and `FitGrid`'s, also when slicing `FitGrid`'s
- fix issue with `plot_betas` used on a grid with a single channel
- add colorscale to `plot_betas`

Features

- Add multicore processing, made possible by making NumPy single-threaded
- Add support for full linear mixed models using `lme4`, replacing `statsmodels`

Bug fixes

- Use all channels by default in `run_model`
- Fix formatting of duplicate epoch indices error message

Other

- Model running is now done sample by sample instead of channel by channel
- Correctness tests for `lm` added

Features

- Add support for mixed models to the extent allowed by `statsmodels`
- Add `Epochs.distances`, a function assigning distances from the "mean epoch" to all epochs. This allows for preliminary outlier rejection.

Bug fixes

- Fix bad `_snapshots` access using `get_group` (38)

Other

- Private `Epochs` attributes are prefixed with `_` to get them out of the way of the user.
- Refactor LHS and RHS validation (DRY).

Features

- Helpful message on duplicate indices (suggested by Melissa)
- `run_model` function in `Epochs` allowing for running arbitrary models
- One shot R^2 plot with marginal subplots
- Averages plotted negative up by default to respect the convention
- Add correct epoch id resolution for returned dataframes, this matters when we get bare arrays from `statsmodels`
- Both `Epochs` and `FitGrid` are channel aware (have a `channels` attribute tracking what channels they have)

Bug fixes
- Font sizes for the plots are changed using context managers so the changes are contained
- `table` inside epochs is now a copy not a reference to the original
- Use Agg matplotlib backend to prevent fatal erros when $DISPLAY is not set
- Copy when creating smaller grids on slicing

Other
- Add API reference to docs
- Trimmed down dependencies
- Better test coverage
- Clearer example in demo notebook